1 Introduction
Hypertension is a major risk factor for ischemic heart disease, stroke, and chronic kidney disease. The global burden of these diseases increased substantially between 1990 and 2010.[1] In 2014, approximately 22% of adults aged ≥18 years had been diagnosed with hypertension worldwide.[2]
Obesity is a major risk factor for hypertension.[3–5] In the Framingham study, weight loss of 6.8 kg or more over 4 years led to a 21% to 29% reduction in hypertension risk.[6] Chandra et al showed that a higher body mass index (BMI) and visceral adiposity were significantly associated with incident hypertension in African–American participants.[7] Lee et al observed that obesity is associated with an increased risk of hypertension in the Korean population, regardless of the presence of elements of metabolic syndrome.[8] However, conventional observational analyses cannot avoid unmeasured confounding and reverse causation, which make it difficult to infer causality from the observed association.[9,10]
Randomized clinical trials (RCTs) have demonstrated the effect of weight loss on blood pressure (BP).[11] However, some RCTs have yielded mixed results. Tyson et al. found that the weight-gain group (>3%) and the weight-stable (within 3%) group both had increased systolic BP (SBP) and that these 2 groups were not significantly different in SBP.[12] Moreover, SBP was unchanged for the weight-loss group who lost 3% or more of their weight. Furthermore, most of the RCTs were short-term studies with small numbers of participants; therefore, the results may not be applicable to the general population and cannot address the long-term health effects of obesity. In addition, intervention could also affect other pathways. For example, weight-loss surgery (eg, Roux-en-Y gastric bypass, laparoscopic adjustable gastric banding, or vertical sleeve gastrectomy) influences glucose metabolism more than it influences the obesity-hypertension pathway.[13]
Mendelian randomization (MR) analysis using genetic variants as the instrumental variable (IV) has been increasingly used to assess causality. Genetic variants are present from conception, allocated randomly according to Mendel second law and are inherited independent of potential confounding factors.[9,10] Thus, the IV (genetic variants associated with obesity) is independent of confounders in its effects on the phenotype (obesity)–outcome (hypertension) relationship.
Recently, a small number of MR studies have reported that BMI has a causal relationship with hypertension.[14–16] However, these studies were conducted in Western populations. The World Health Organization reported that the prevalence of overweight (BMI ≥ 25 kg/m2) and obesity (BMI ≥ 30 kg/m2) is highest in the Americas (61% overweight or obese in both sexes, and 27% obese), especially in the United States (68% overweight or obese among both sexes, and 32% obese). In contrast, Koreans have a low prevalence of obesity (31% overweight or obese among both sexes and 4.6% obese). However, the prevalence of hypertension is similar between the United States and Korea (9.4% vs 8.4%, respectively).[2,17] Because of the different prevalences of obesity but similar prevalences of hypertension between the United States and Korea, a study of the causal relationship between obesity and hypertension in Korea is needed.
A composite genetic risk score (GRS) could reduce the statistical error associated with multiple testing compared to individual single-nucleotide polymorphisms (SNPs). Thus, we analyzed the association between IV for obesity using BMI-associated GRS (BMI GRS) and risk of hypertension to explore the causal association between obesity and hypertension.
2 Methods
2.1 Study population
We used data from the Ansung-Ansan cohort within the Korean Genome and Epidemiology Study (KoGES), which was initiated in 2001 as a population-based cohort study recruiting Korean adults aged 40 to 69 years. Briefly, a total of 5020 participants (2523 men and 2497 women) in Ansan and 5018 participants (2239 men and 2779 women) in Ansung were included in the baseline examinations from June 2001 to January 2003. Follow-up surveys were conducted biennially, and study participants were followed-up to 5 times until 2012. Information about their general characteristics, lifestyle, and current medications was obtained through questionnaires. Physical examinations, including BP, anthropometric measurements, and blood sampling, were conducted by trained researchers from 2001 to 2012. During this 10-year period, a follow-up rate of 62.1% was achieved.
The criteria for exclusion were no data of genotype (n = 1196), missing BP measurements, or history of hypertension diagnosis (n = 10). After this exclusion, the present report focuses on 8832 participants for whom information about the genotype and outcome variables of hypertension were available.
An informed consent form was signed by each participant, and the study protocol was approved by the institutional review board of the Seoul National University Hospital (IRB No: 1312-033-539).
2.2 Genotype
A total of 10,004 participants were genotyped using the Affymetrix Genome-Wide Human SNP Array 5.0 (Santa Clara, CA) containing 500,568 SNPs. Genotype clustering was determined using Bayesian robust linear modeling of the Mahalanobis distance. Before statistical analysis, 17,926 markers with a genotype call rate <95%, 92,050 markers with low minor allele frequency (<0.01), and 38,364 markers with Hardy–Weinberg equilibrium (P < 10−6) were removed, leaving 352,228 SNPs for 8842 individuals. An additional 1.8 × 106 SNPs were found by imputation using the JPT/CHB component of HapMap as the reference. After filtering, a total of 1,590,162 genotyped and imputed SNPs were available for analyses. The genotyping methods of the KoGES have been described previously in detail.[18]
2.3 BMI and lifestyle measurement
Alcohol consumption was calculated as the amount consumed per week and divided into 2 groups. Based on the guidelines for recommended alcohol consumption to lower health risks from the Korea Health Promotion Foundation, we defined low consumption of alcohol as 40 g or less for males and 20 g or less for females at one time, less than twice a week.[19] They were also split into 2 groups by smoking status: <20 pack-years smoking and >20 pack-years. BMI was calculated as weight in kilograms divided by height in meters squared (kg/m2) at the baseline survey.
2.4 Hypertension assessment
The BP was measured using mercury sphygmomanometers (Baumanometer; WA Baum, Copiague, NY) according to a standardized protocol.[20] All measurements in the present study were taken after at least a 5-min rest. We used an average of 3 measurements. At baseline, hypertensive participants were defined as having SBP ≥ 140 mm Hg or diastolic BP (DBP) ≥ 90 mm Hg, using antihypertensive medication, or having a history of hypertension diagnosed by a doctor. After these participants were excluded, newly diagnosed cases of hypertension were defined as SBP ≥ 140 or DBP ≥ 90 mm Hg and taking antihypertensive drugs during the 10-year follow-up. We considered both baseline hypertension and newly diagnosed hypertension during the 10-year follow-up period.
2.5 Selection of genetic loci and GRS construction
We selected individual 32 SNPs associated with BMI using linear regression (Supplemental Table S1, https://links.lww.com/MD/C348). Because we wanted to include more SNPs, we used a liberal P-value (<1.0 × 10−5) instead of a restrictive P-value after Bonferroni correction, 5.0 × 10−8. Among these SNPs, 2 SNPs had been reported previously.[18] Some SNPs were found to be in high linkage disequilibrium (|D′| ≥ 0.9). Therefore, we selected 1 representative SNP from the closely linked SNPs based on the estimated size of the main genetic analysis results or significance in previous studies. Finally, 3 BMI GRSs were constructed. The first BMI GRS was composed of 2 significant SNPs (rs17178527 and rs9939609) found in a previous study.[18] The second was composed of 4 SNPs (rs17178527, rs9939609, rs7668087, and rs11000212) selected with a cut-off P-value <5 × 10−6. The third composed of 6 SNPs (rs17178527, rs9939609, rs7668087, rs11000212, rs17130257, and rs10936246) selected with a cut-off P-value <5 × 10−5. The GRS was produced by 2 methods: a simple-count method (CGRS) and a weighted method (WGRS).[21,22] Six types (3 × 2) of BMI GRS (CGRS [n = 2], WGRS [n = 2], CGRS [n = 4], WGRS [n = 4], CGRS [n = 6], and WGRS [n = 6]) were used in the analysis. We assumed an additive genetic model for each SNP, applying a linear weighting of 0, 1, or 2 to genotypes containing 0, 1, or 2 risk alleles, respectively. The simple-count model assumes that each SNP in the panel contributes equally to the risk of hypertension and was calculated by summing the values (0, 1, and 2) for each of the SNPs. The weighted GRS was calculated by multiplying each β coefficient obtained from linear regression by the number of corresponding risk alleles (0, 1, and 2). All β coefficients were positive because we reordered the sequence of genotypes of the SNPs when the weights were less than zero.
2.6 Statistical analysis
Multivariate logistic regression models were used to assess the association between BMI and hypertension. Model 1 was not adjusted for other variables; Model 2 was adjusted for age (years) and sex (male or female); Model 3 was further adjusted for region (Ansung or Ansan), education (≤9 or >9 years of school), tobacco smoking, and current alcohol consumption. The association between BMI GRS and hypertension was evaluated in a bivariate logistic regression model. In MR analysis, we used the 6 types of BMI GRS as the IV estimators to measure the strength of the causal relationship between BMI and hypertension. The IV estimate of causal odds ratio (OR) was derived using the Wald-type estimator and then exponentiation to express the result as an OR.[14] ORGRS-hypertension estimated the effect of the GRS on hypertension using univariate logistic regression. βGRS-BMI estimated the effect of the GRS on BMI using linear regression.
;)
We also tested the difference between the IV estimators and the conventional regression-based estimators for the effect of BMI using a classical z test. Figure 1 shows the directed acyclic graphs between exposure (BMI) and outcome (hypertension) with the genetic instrument.
Figure 1:
Directed acyclic graph explaining the relationships between exposure (BMI) and outcome (hypertension) with the genetic instrument (genetic score). BMI = body mass index, CGRS = count genetic risk score, IV = instrumental variable, OR = odds ratio.
In the sensitivity analysis, we conducted MR analysis using only baseline data for a cross-sectional approach. Statistical significance was set to a 2-sided P-value of <.05. All statistical analyses were performed using SAS software version 9.3 (SAS Institute Inc, Cary, NC), Plink (version 1.08, http://pngu.mgh.harvard.edu/∼purcell/plink), and R version 3.1.0 (Comprehensive R Archive Network: http://cran.r-project.org).
3 Results
Among 8832 participants, 4179 (47.3%) were men. The average age was 52 (SD 8.92) years, and the average BMI was 24.6 (SD 3.12) kg/m2. At baseline, hypertension was diagnosed in 2971 (33.6%) participants, and the remaining 5861 (66.4%) participants were not hypertensive. During the 10-year follow-up, hypertension was newly detected in 1409 participants (first follow-up: 436; second follow-up: 274; third follow-up: 232; fourth follow-up: 322; and fifth follow-up: 145) (Supplemental Fig. S1, https://links.lww.com/MD/C348). Therefore, the number (proportion) of hypertension at baseline and new cases during follow-up was 4380 (49.6%) and no hypertension during follow-up was 4452 (50.4%). As shown in Table 1, there were statistically significant differences in age, area, education, alcohol consumption, and BMI measured between hypertension at baseline/follow-up and no hypertension during follow-up. Table 2 shows the demographic features of the participants according to BMI GRS quartiles using baseline and longitudinal data. The BMI GRS (in quartiles) was significantly associated with BMI (P for trend <.0001). No other population characteristics (sex, area, smoking, current alcohol drinking) were associated with the BMI GRS (n = 6) quartiles (all P for trend >.05).
Table 1:
General characteristics of the study population using baseline and longitudinal data (n = 8832).
Table 2:
Characteristics of study participants according to the weighted BMI genetic risk score (BMI GRS) using baseline and longitudinal data.
As shown in Table 3, in the multivariable adjusted model using baseline and longitudinal data, each 1 kg/m2 increase in BMI was associated with an 19% (OR: 1.19, 95% confidence interval [CI]: 1.17–1.21) increase in hypertension risk. In the multivariable adjusted model, each SD increase in the 6 types of BMI-GRS was associated with a 5% to 6% (OR: 1.05–1.06) increase in the risk of hypertension except for WGRS (n = 6).
Table 3:
The association of BMI GRS and BMI with hypertension using baseline and longitudinal data.
Figure 2 shows the MR results using baseline and longitudinal data. In the IV analysis, the causal OR of a 1 kg/m2 increase in BMI for hypertension was 1.13 to 1.26 (all P-value <.05, except WGRS [n = 6]).Compared to the IV using GRS (n = 4 or 6), IV using GRS (n = 2) yielded a greater OR in MR analysis. The causal estimate of the relationship between BMI and hypertension risk using the IV variable and the observed association between BMI and hypertension risk were not significantly different in a classical z-test (1.13–1.26 vs 1.19, P > .05).
Figure 2:
Instrumental variable-estimated association of body mass index and hypertension (baseline and newly diagnosed hypertension) using baseline and longitudinal data. CGRS = count genetic risk score, CI = confidence interval, OR = odds ratio, WGRS = weighted genetic risk score.
We also conducted a sensitivity analysis with baseline hypertension only. Supplemental Table S2, https://links.lww.com/MD/C348 showed that each SD increase in CGRS (n = 2) and WGRS (n = 2) was associated with a 6% (OR: 1.06) increased risk of baseline hypertension in the multivariable adjusted model using baseline data only. Each 1 kg/m2 increase in BMI was associated with a 18% (OR: 1.18, 95% CI: 1.17–1.20) increased baseline hypertension risk in multivariable analysis using baseline data only. In the IV analysis, BMI was found to have a causal relationship with baseline hypertension for CGRS (n = 2) and WGRS (n = 2) using baseline data only. The causal OR of a 1 kg/m2 increase in BMI for hypertension was 1.25 and 1.26 (P < .05) (Supplemental Fig. S2, https://links.lww.com/MD/C348), and there was no significant difference between IV analysis and multivariate analysis in a classical z test (1.25–1.26 vs 1.18, P > .05) using baseline data only.
4 Discussion
Using the data from a 10-year follow-up investigation, including 8832 community-dwelling Korean middle-aged adults, we performed an analysis utilizing an MR design and provided additional evidence to support the causal role of BMI in hypertension. These findings are consistent with evidence from observational studies that have demonstrated the association of high BMI with increased risk of hypertension.[8] This evidence provides a rationale to further investigate whether weight-control programs can reduce the incidence of hypertension in those who are at risk.
Several pathogenic mechanisms have been suggested to contribute to the development of hypertension in an obese population: insulin resistance, vascular alterations, and activation of the renin-angiotensin-aldosterone system.[23,24] Excess adipocyte tissue stimulates insulin secretion, which activates the sympathetic nervous system and raises the BP.[25] Insulin also acts directly on the kidneys to stimulate sodium retention, increase plasma volume, and raise the BP.[26] Vascular alterations, including structural changes, endothelial dysfunction, and altered stiffness, are common in obesity and are also thought to contribute to the development of hypertension.[27,28] An activated renin-angiotensin-aldosterone system in the excess adipose tissue of obese people generates angiotensin and aldosterone, which again elevate the BP.[23]
An important difference between conventional RCTs and MR studies using genetic polymorphisms is that MR studies evaluated the association between lifetime exposure to selected alleles in the general population with an outcome, whereas conventional RCTs provide insights for shorter periods among more selected individuals.[29]
Previously, a small number of MR studies have provided evidence supporting a causal link between BMI and hypertension. Fall et al demonstrated a significant association between the adiposity-associated variant rs9939609 at the FTO locus and SBP and suggested a possible causal association with elevated SBP (+0.89 mm Hg/[kg/m2]).[14] In this study, rs9939609 at the FTO locus was included in the genetic risk score. Fall et al also constructed a GRS using 32 SNPs and reported a causal effect of adiposity on BP within the European Network for Genetic and Genomic Epidemiology Consortium.[15] Holmes et al performed a genetic-association study of BMI using the CardioChip, used the results to construct a GRS comprising 14 SNPs the group's and showed that a 1 kg/m2 increase in BMI increased SBP by 0.70 mm Hg (95% CI: 0.24–1.16) and DBP by 0.28 mm Hg (95% CI: 0.03–0.52) in the US population.[16]
An MR study is a valid way to explore evidence for causality, given that certain assumptions are met. First, there has to be a strong association between genetic variant (IV) and the exposure of interest. Two SNPs (rs17178527, rs9939609) used in this study have previously been shown to be strongly associated with BMI,[18,30,31] a finding that was replicated in our present study. To assess the relevance of the instruments, we tested the F-statistic in the first-stage regression (IV association with the risk exposure). As a rule of thumb, if the F-statistic was smaller than 10, the IV was defined as a “weak instrument.”[32] In our study, the F-statistics for all BMI GRSs were >10 (52.7–125.3), so problems associated with weak instruments were unlikely. Second, the IV must be independent of covariates. In our study, the IV was independent with measurable covariates (age, sex, area, education, smoking, and alcohol consumption). Third, there are no other pathways between the genetic variant and outcomes (pleiotropy). However, this assumption is untestable. The rs9939609 SNP on the FTO gene has no known pleiotropy.[14] However, the other SNPs were not validated to exclude pleiotropy. Because the quality of evidence provided by an MR study relies heavily on these assumptions,[33] and these MR analyses using 6 different GRSs provided consistent results, although GRS (n = 2) yielded a greater OR than did GRS (n = 4) and GRS (n = 6). This difference might be due to the inclusion of additional marginally significant SNPs, which would reduce the strength and precision of a SNP-exposure association. Likewise, Vassy et al found that 62-SNP GRS did not substantively improve prediction of type-2 diabetes compared with a 40-SNP GRS.[34] More work is needed to determine whether SNPs that do not reach stringent genome-wide significance levels in GRSs should be included in MR studies.
Our main MR analysis considered both prevalent and incident hypertension cases. Additional sensitivity analyses (CGRS [n = 2] and WGRS [n = 2]) using only prevalent cases at baseline also showed a causal effect of adiposity on hypertension.
The strength of the present study is the well-defined community setting and a relatively large sample. To our knowledge, this is the first report showing the effect of common genetic variations related to BMI as the IV in measuring association with hypertension in an East Asian population.
With regard to the limitations of the present study, first, we built the BMI GRS based only on common variants, so we were unable to assess the potential contribution of rare variants. Second, this study did not include any subjects aged ≥70 years or <40 years, so these results may not be generalizable to populations of different ages. Similarly, the results may not be generalizable populations of different ethnicities because we used a cohort composed only of Koreans. Third, this study examined the causal effect of obesity on BP, but we could not test the impact of acute changes. Finally, because we used just a single study, we have limited MR-Egger regression to select candidate SNPs. MR-Egger regression was a 2-sample MR study in which multiple genetic variants affect the outcome.[35,36] Further MR-Egger regression using 2 independent data, particularly of community cohort with genetic information, would be necessary.
5 Conclusion
We found that genetic predisposition for a higher BMI was associated with higher risk of hypertension in the Korean population. This MR analysis provided evidence of a causal relationship between BMI and hypertension. Our results suggest that controlling obesity may be beneficial for the prevention of hypertension.
Author contributions
Conceptualization: Yun-Chul Hong.
Methodology: Youn-Hee Lim.
Project administration: Youn-Hee Lim.
Supervision: Yun-Chul Hong.
Writing – original draft: Mee-Ri Lee.
Writing – review & editing: Mee-Ri Lee.
References
[1]. Lim SS, Vos T, Flaxman AD, et al. A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990-2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet 2012;380:2224–60.
[2]. WHO. Global status report on noncommunicable diseases 2014. Attaining the nine global noncommunicable diseases targets; a shared responsibility 2014;
http://www.who.int/nmh/publications/ncd-status-report-2014/en/. Accessed 11. July, 2018.
[3]. Ezzati M, Riboli E. Behavioral and dietary risk factors for noncommunicable diseases. N Eng J Med 2013;369:954–64.
[4]. Stokes J 3rd, Kannel WB, Wolf PA, et al. Blood pressure as a risk factor for cardiovascular disease. The Framingham Study - 30 years of follow-up. Hypertension 1989;13(Suppl):I13–8.
[5]. Kannel WB. Fifty years of Framingham Study contributions to understanding hypertension. J Hum Hypertens 2000;14:83–90.
[6]. Moore LL, Visioni AJ, Qureshi MM, et al. Weight loss in overweight adults and the long-term risk of hypertension: the Framingham study. Archives of internal medicine 2005;165:1298–303.
[7]. Chandra A, Neeland IJ, Berry JD, et al. The relationship of body mass and fat distribution with incident hypertension: observations from the Dallas Heart Study. J Am Coll Cardiol 2014;64:997–1002.
[8]. Lee SK, Kim SH, Cho GY, et al. Obesity phenotype and incident hypertension: a prospective community-based cohort study. J Hypertens 2013;31:145–51.
[9]. Smith GD. ESMRGVaIfSCIiOSINRCUCoAiCa. Biosocial Surveys. Washington, DC: National Academies Press; 2008.
[10]. Smith GD, Ebrahim S. Mendelian randomization: prospects, potentials, and limitations. Int J Epidemiol 2004;33:30–42.
[11]. Horvath K, Jeitler K, Siering U, et al. Long-term effects of weight-reducing interventions in hypertensive patients: systematic review and meta-analysis. Arch Intern Med 2008;168:571–80.
[12]. Tyson CC, Appel LJ, Vollmer WM, et al. Impact of 5-year weight change on blood pressure: results from the Weight Loss Maintenance trial. J Clin Hypertens (Greenwich) 2013;15:458–64.
[13]. Madsbad S, Dirksen C, Holst JJ. Mechanisms of changes in glucose metabolism and bodyweight after bariatric surgery. Lancet Diabetes Endocrinol 2014;2:152–64.
[14]. Fall T, Hagg S, Magi R, et al. The role of adiposity in cardiometabolic traits: a Mendelian randomization analysis. PLoS Med 2013;10:e1001474.
[15]. Fall T, Hagg S, Ploner A, et al. Age- and sex-specific causal effects of adiposity on cardiovascular risk factors. Diabetes 2015;64:1841–52.
[16]. Holmes MV, Lange LA, Palmer T, et al. Causal effects of body mass index on cardiometabolic traits and events: a Mendelian randomization analysis. Am J Hum Genet 2014;94:198–208.
[18]. Cho YS, Go MJ, Kim YJ, et al. A large-scale genome-wide association study of Asian populations uncovers genetic factors influencing eight quantitative traits. Nat Genet 2009;41:527–34.
[19]. KHPF. Low Alcohol Consumption Guideline. 2013;Seoul, Korea: KHPF, 5.
[20]. 1999 World Health Organization-International Society of Hypertension Guidelines for the Management of Hypertension. Guidelines Subcommittee. J Hypertens 1999;17:151–83.
[21]. Cornelis MC, Qi L, Zhang C, et al. Joint effects of common genetic variants on the risk for type 2 diabetes in U.S. men and women of European ancestry. Ann Intern Med 2009;150:541–50.
[22]. Ripatti S, Tikkanen E, Orho-Melander M, et al. A multilocus genetic risk score for coronary heart disease: case-control and prospective cohort analyses. Lancet 2010;376:1393–400.
[23]. Rahmouni K. Obesity-associated hypertension: recent progress in deciphering the pathogenesis. Hypertension 2014;64:215–21.
[24]. Vaneckova I, Maletinska L, Behuliak M, et al. Obesity-related hypertension: possible pathophysiological mechanisms. J Endocrinol 2014;223:R63–78.
[25]. Landsberg L, Aronne LJ, Beilin LJ, et al. Obesity-related hypertension: pathogenesis, cardiovascular risk, and treatment: a position paper of The Obesity Society and the American Society of Hypertension. J Clin Hyperten (Greenwich) 2013;15:14–33.
[26]. DeFronzo RA. Insulin and renal sodium handling: clinical implications. Intern J Obesity 1981;5(Suppl 1):93–104.
[27]. De Ciuceis C, Porteri E, Rizzoni D, et al. Effects of weight loss on structural and functional alterations of subcutaneous small arteries in obese patients. Hypertension 2011;58:29–36.
[28]. Weisbrod RM, Shiang T, Al Sayah L, et al. Arterial stiffening precedes systolic hypertension in diet-induced obesity. Hypertension 2013;62:1105–10.
[29]. Mokry LE, Ross S, Ahmad OS, et al. Vitamin D and risk of multiple sclerosis: a Mendelian randomization study. PLoS Med 2015;12:e1001866.
[30]. Frayling TM, Timpson NJ, Weedon MN, et al. A common variant in the FTO gene is associated with body mass index and predisposes to childhood and adult obesity. Science 2007;316:889–94.
[31]. Willer CJ, Speliotes EK, Loos RJ, et al. Six new loci associated with body mass index highlight a neuronal influence on body weight regulation. Nat Genet 2009;41:25–34.
[32]. Burgess S, Thompson SG, Collaboration CCG. Avoiding bias from weak instruments in Mendelian randomization studies. Int J Epidemiol 2011;40:755–64.
[33]. Burgess S, Butterworth AS, Thompson JR. Beyond Mendelian randomization: how to interpret evidence of shared genetic predictors. J Clin Epidemiol 2016;69:208–16.
[34]. Vassy JL, Hivert MF, Porneala B, et al. Polygenic type 2 diabetes prediction at the limit of common variant detection. Diabetes 2014;63:2172–82.
[35]. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol 2015;44:512–25.
[36]. Bowden J. Misconceptions on the use of MR-Egger regression and the evaluation of the InSIDE assumption. Int J Epidemiol 2017;46:2097–9.
